Dialog Acts from the Processing Perspective in Task Oriented Dialog Systems
Markus M. Berg, Bernhard Thalheim · 2011
The formulation ”I’d like to know what time it is” has the same aim as ”What’s the time?”. Thus, we can easily see that different formulations can have the same intention. Consequently we learn that it is not possible to infer a one-to-one relationship between form and function. When developing a dialogue system, the main interest is what the user expects from the system, and not how he formulates his concern. So we propose a backendoriented scheme for the description of dialog utterances. This scheme applies for three basic types of mixed-initiative systems that often have to be modeled: control systems (e.g. for controlling the lights in a room by speech), question-answering systems and information-seeking/booking-dialogues (i.e. the system asks questions in order to gain information that is necessary to fulfil the user’s request). All of these systems have a task-related information exchange in common. Thus we don’t classify by initiative (they are all mixed initiative) but by purpose and call those systems, according to (McTear, 2004, p.45), ”task-oriented” dialog systems.